Type safe Django app, Part 2

Let's run our initial migrations and get rid of the error that we ignored in the first part.

1

$ python manage.py migrate

Create models for our Polls app. Nothing new here, but notice the nice type hits you get for all the fields thanks to django-stubs.

polls/models.py:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

from django.db import models


class Question(models.Model):
    question_text = models.CharField(max_length=200)
    pub_date = models.DateTimeField('date published')


class Choice(models.Model):
    question = models.ForeignKey(Question, on_delete=models.CASCADE)
    choice_text = models.CharField(max_length=200)
    votes = models.IntegerField(default=0)

Before we can apply these migrations, we need to let Django know about our Polls app.

django_returns/settings.py:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14

...

INSTALLED_APPS = [
    # add this
    'polls.apps.PollsConfig',
    'django.contrib.admin',
    'django.contrib.auth',
    'django.contrib.contenttypes',
    'django.contrib.sessions',
    'django.contrib.messages',
    'django.contrib.staticfiles',
]

...

Now, we need to generate and run the migrations for our new models:

1
2

$ python manage.py makemigrations polls
$ python manage.py migrate

We will start a little out of order by improving was_published_recently. Since __str__ is a magic method, it will require a little extra work and build on what we will learn when we write was_published_recently.

polls/models.py:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11

import datetime

from django.db import models
from django.utils import timezone
from returns.io import impure_safe

class Questions(models.Model):
    # ...
    @impure_safe
    def was_published_recently(self) -> bool:
        return self.pub_date >= timezone.now() - datetime.timedelta(days=1)

First off, note that when we write methods, we do not add a type to self. We do, however, add a return type to all methods. This one lets us know that we will get back a boolean value.

We also added the impure_safe decorator. As the name implies, this decorator helps us mark functions that are impure and unsafe. was_published_recently is impure because it reaches out to the database, where we can't guarantee what we get back, and relies on time, which constantly changes.

was_published_recently is also unsafe because of the dependency on the database. All kinds of things can go wrong when we try to get a value out of a database: the database could be down, the table might not exist, the model fields could have changed, we forgot to run our migrations, etc.

impure_safe doesn't fix these issues, but it lets the caller know that this is no longer a simple function that returns a bool.

If we try to call this functions, we will see that impure_safe changed our bool return type to IOResultE[bool, Exception]. This wrapper puts two layers around the bool value. The first layer, Result, prevents us from getting the value without first checking and handling the Exception. The second layer, IO, prevents us from using the value outside of other IO functions (we don't want to mix pure and impure code).

Next, we will start our implementation __str__. As mentioned earlier, magic methods are a little tricky to handle. It would be nice if we could just annotate __str__ with impure_safe, but python won't know how to handle our IOResultE wrapper. So we have to come to a compromise.

polls/models.py:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19

#...

from returns.unsafe import unsafe_perform_io


class Question(models.Model):
    #...

    @impure_safe
    def safe_str(self) -> str:
        return self.question_text


class Choice(models.Model):
    #...

    @impure_safe
    def safe_str(self) -> str:
        return self.choice_text

We create a safe_str function with the impure_safe decorator. When we need to get a string representation in our code, we should default to using this function since it has all of our safety improvements.

safe_str helps our code, but we will still need a proper __str__ implementation so we can get nice output at our shell.

Since this functionality is mainly for the shell, this is a good time to fire up a shell session. If we run safe_str or was_published_recentrly, we will notice that the return value is <IOResult: <Success: ...>> this is the printable representation of the two wrappers.

We can take off the Result wrapper by running value_or("error message"). This gives us back a <IO: ...> value. Unlike Result which we can unwrap and inspect in a safe manner (as long as we handle any failures), IO is always unsafe. In order to remove the IO wrapper we need to run the ominously named unsafe_perform_io.

We should be very careful with unsafe_perform_io. Running unsafe_perform_io removes all the safety that we worked so hard building up. But because python is, by its nature, unsafe we will need to use it. In order to keep as much safety as possible, however, we must limit its usage to the places where we need to hand off the value to python, Django, or any library outside our control. The __str__ magic method is exactly this kind of place, so it's ok to use unsafe_perform_io there.

The use of unsafe_perform_io on the edges of our application is the basis of a common design patter known as imperative shell and functional core. returns automatically pushes us towards this pattern, and we will see how to further utilize it in the next part.

Now that we know how to remove the our wrappers, we can implement __str__ in the safest possible manner.

polls/models.py:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

#...

class Question(models.Model):
    #...

    def __str__(self) -> str:
        value = self.safe_str().value_or("No question_text found")
        return unsafe_perform_io(value)


class Choice(models.Model):
    #...

    def __str__(self) -> str:
        value = self.safe_str().value_or("No choice_text available")
        return unsafe_perform_io(value)

This part of the Django tutorial introduced the simplest way to work with returns. The benefits seem marginal, if any for now, but the real power of returns and this type safe approach comes throw when we need to compose multiple functions with different wrappers and properties. That will be the focus of the next part of this tutorial.

In the meantime, I encourage you to go through at least all the shell examples in the official tutorial and explore how our changes made these functions work differently than what you see in the tutorial.